2 * Copyright (c) 2000-2003 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_format.h"
21 #include "xfs_shared.h"
24 #include "xfs_trans.h"
27 #include "xfs_alloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_inode.h"
33 #include "xfs_bmap_util.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_error.h"
36 #include "xfs_itable.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_trans_priv.h"
42 #include "xfs_cksum.h"
43 #include "xfs_trace.h"
50 * dquot->q_qlock (xfs_dqlock() and friends)
51 * dquot->q_flush (xfs_dqflock() and friends)
54 * If two dquots need to be locked the order is user before group/project,
55 * otherwise by the lowest id first, see xfs_dqlock2.
59 xfs_buftarg_t
*xfs_dqerror_target
;
62 int xfs_dqerror_mod
= 33;
65 struct kmem_zone
*xfs_qm_dqtrxzone
;
66 static struct kmem_zone
*xfs_qm_dqzone
;
68 static struct lock_class_key xfs_dquot_group_class
;
69 static struct lock_class_key xfs_dquot_project_class
;
72 * This is called to free all the memory associated with a dquot
78 ASSERT(list_empty(&dqp
->q_lru
));
80 mutex_destroy(&dqp
->q_qlock
);
81 kmem_zone_free(xfs_qm_dqzone
, dqp
);
83 XFS_STATS_DEC(xs_qm_dquot
);
87 * If default limits are in force, push them into the dquot now.
88 * We overwrite the dquot limits only if they are zero and this
89 * is not the root dquot.
92 xfs_qm_adjust_dqlimits(
96 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
97 struct xfs_disk_dquot
*d
= &dq
->q_core
;
102 if (q
->qi_bsoftlimit
&& !d
->d_blk_softlimit
) {
103 d
->d_blk_softlimit
= cpu_to_be64(q
->qi_bsoftlimit
);
106 if (q
->qi_bhardlimit
&& !d
->d_blk_hardlimit
) {
107 d
->d_blk_hardlimit
= cpu_to_be64(q
->qi_bhardlimit
);
110 if (q
->qi_isoftlimit
&& !d
->d_ino_softlimit
)
111 d
->d_ino_softlimit
= cpu_to_be64(q
->qi_isoftlimit
);
112 if (q
->qi_ihardlimit
&& !d
->d_ino_hardlimit
)
113 d
->d_ino_hardlimit
= cpu_to_be64(q
->qi_ihardlimit
);
114 if (q
->qi_rtbsoftlimit
&& !d
->d_rtb_softlimit
)
115 d
->d_rtb_softlimit
= cpu_to_be64(q
->qi_rtbsoftlimit
);
116 if (q
->qi_rtbhardlimit
&& !d
->d_rtb_hardlimit
)
117 d
->d_rtb_hardlimit
= cpu_to_be64(q
->qi_rtbhardlimit
);
120 xfs_dquot_set_prealloc_limits(dq
);
124 * Check the limits and timers of a dquot and start or reset timers
126 * This gets called even when quota enforcement is OFF, which makes our
127 * life a little less complicated. (We just don't reject any quota
128 * reservations in that case, when enforcement is off).
129 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
131 * In contrast, warnings are a little different in that they don't
132 * 'automatically' get started when limits get exceeded. They do
133 * get reset to zero, however, when we find the count to be under
134 * the soft limit (they are only ever set non-zero via userspace).
137 xfs_qm_adjust_dqtimers(
144 if (d
->d_blk_hardlimit
)
145 ASSERT(be64_to_cpu(d
->d_blk_softlimit
) <=
146 be64_to_cpu(d
->d_blk_hardlimit
));
147 if (d
->d_ino_hardlimit
)
148 ASSERT(be64_to_cpu(d
->d_ino_softlimit
) <=
149 be64_to_cpu(d
->d_ino_hardlimit
));
150 if (d
->d_rtb_hardlimit
)
151 ASSERT(be64_to_cpu(d
->d_rtb_softlimit
) <=
152 be64_to_cpu(d
->d_rtb_hardlimit
));
156 if ((d
->d_blk_softlimit
&&
157 (be64_to_cpu(d
->d_bcount
) >
158 be64_to_cpu(d
->d_blk_softlimit
))) ||
159 (d
->d_blk_hardlimit
&&
160 (be64_to_cpu(d
->d_bcount
) >
161 be64_to_cpu(d
->d_blk_hardlimit
)))) {
162 d
->d_btimer
= cpu_to_be32(get_seconds() +
163 mp
->m_quotainfo
->qi_btimelimit
);
168 if ((!d
->d_blk_softlimit
||
169 (be64_to_cpu(d
->d_bcount
) <=
170 be64_to_cpu(d
->d_blk_softlimit
))) &&
171 (!d
->d_blk_hardlimit
||
172 (be64_to_cpu(d
->d_bcount
) <=
173 be64_to_cpu(d
->d_blk_hardlimit
)))) {
179 if ((d
->d_ino_softlimit
&&
180 (be64_to_cpu(d
->d_icount
) >
181 be64_to_cpu(d
->d_ino_softlimit
))) ||
182 (d
->d_ino_hardlimit
&&
183 (be64_to_cpu(d
->d_icount
) >
184 be64_to_cpu(d
->d_ino_hardlimit
)))) {
185 d
->d_itimer
= cpu_to_be32(get_seconds() +
186 mp
->m_quotainfo
->qi_itimelimit
);
191 if ((!d
->d_ino_softlimit
||
192 (be64_to_cpu(d
->d_icount
) <=
193 be64_to_cpu(d
->d_ino_softlimit
))) &&
194 (!d
->d_ino_hardlimit
||
195 (be64_to_cpu(d
->d_icount
) <=
196 be64_to_cpu(d
->d_ino_hardlimit
)))) {
201 if (!d
->d_rtbtimer
) {
202 if ((d
->d_rtb_softlimit
&&
203 (be64_to_cpu(d
->d_rtbcount
) >
204 be64_to_cpu(d
->d_rtb_softlimit
))) ||
205 (d
->d_rtb_hardlimit
&&
206 (be64_to_cpu(d
->d_rtbcount
) >
207 be64_to_cpu(d
->d_rtb_hardlimit
)))) {
208 d
->d_rtbtimer
= cpu_to_be32(get_seconds() +
209 mp
->m_quotainfo
->qi_rtbtimelimit
);
214 if ((!d
->d_rtb_softlimit
||
215 (be64_to_cpu(d
->d_rtbcount
) <=
216 be64_to_cpu(d
->d_rtb_softlimit
))) &&
217 (!d
->d_rtb_hardlimit
||
218 (be64_to_cpu(d
->d_rtbcount
) <=
219 be64_to_cpu(d
->d_rtb_hardlimit
)))) {
226 * initialize a buffer full of dquots and log the whole thing
229 xfs_qm_init_dquot_blk(
236 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
241 ASSERT(xfs_buf_islocked(bp
));
246 * ID of the first dquot in the block - id's are zero based.
248 curid
= id
- (id
% q
->qi_dqperchunk
);
250 memset(d
, 0, BBTOB(q
->qi_dqchunklen
));
251 for (i
= 0; i
< q
->qi_dqperchunk
; i
++, d
++, curid
++) {
252 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
253 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
254 d
->dd_diskdq
.d_id
= cpu_to_be32(curid
);
255 d
->dd_diskdq
.d_flags
= type
;
256 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
257 uuid_copy(&d
->dd_uuid
, &mp
->m_sb
.sb_uuid
);
258 xfs_update_cksum((char *)d
, sizeof(struct xfs_dqblk
),
263 xfs_trans_dquot_buf(tp
, bp
,
264 (type
& XFS_DQ_USER
? XFS_BLF_UDQUOT_BUF
:
265 ((type
& XFS_DQ_PROJ
) ? XFS_BLF_PDQUOT_BUF
:
266 XFS_BLF_GDQUOT_BUF
)));
267 xfs_trans_log_buf(tp
, bp
, 0, BBTOB(q
->qi_dqchunklen
) - 1);
271 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
272 * watermarks correspond to the soft and hard limits by default. If a soft limit
273 * is not specified, we use 95% of the hard limit.
276 xfs_dquot_set_prealloc_limits(struct xfs_dquot
*dqp
)
280 dqp
->q_prealloc_hi_wmark
= be64_to_cpu(dqp
->q_core
.d_blk_hardlimit
);
281 dqp
->q_prealloc_lo_wmark
= be64_to_cpu(dqp
->q_core
.d_blk_softlimit
);
282 if (!dqp
->q_prealloc_lo_wmark
) {
283 dqp
->q_prealloc_lo_wmark
= dqp
->q_prealloc_hi_wmark
;
284 do_div(dqp
->q_prealloc_lo_wmark
, 100);
285 dqp
->q_prealloc_lo_wmark
*= 95;
288 space
= dqp
->q_prealloc_hi_wmark
;
291 dqp
->q_low_space
[XFS_QLOWSP_1_PCNT
] = space
;
292 dqp
->q_low_space
[XFS_QLOWSP_3_PCNT
] = space
* 3;
293 dqp
->q_low_space
[XFS_QLOWSP_5_PCNT
] = space
* 5;
297 xfs_dquot_buf_verify_crc(
298 struct xfs_mount
*mp
,
301 struct xfs_dqblk
*d
= (struct xfs_dqblk
*)bp
->b_addr
;
305 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
309 * if we are in log recovery, the quota subsystem has not been
310 * initialised so we have no quotainfo structure. In that case, we need
311 * to manually calculate the number of dquots in the buffer.
314 ndquots
= mp
->m_quotainfo
->qi_dqperchunk
;
316 ndquots
= xfs_qm_calc_dquots_per_chunk(mp
,
317 XFS_BB_TO_FSB(mp
, bp
->b_length
));
319 for (i
= 0; i
< ndquots
; i
++, d
++) {
320 if (!xfs_verify_cksum((char *)d
, sizeof(struct xfs_dqblk
),
323 if (!uuid_equal(&d
->dd_uuid
, &mp
->m_sb
.sb_uuid
))
330 xfs_dquot_buf_verify(
331 struct xfs_mount
*mp
,
334 struct xfs_dqblk
*d
= (struct xfs_dqblk
*)bp
->b_addr
;
340 * if we are in log recovery, the quota subsystem has not been
341 * initialised so we have no quotainfo structure. In that case, we need
342 * to manually calculate the number of dquots in the buffer.
345 ndquots
= mp
->m_quotainfo
->qi_dqperchunk
;
347 ndquots
= xfs_qm_calc_dquots_per_chunk(mp
, bp
->b_length
);
350 * On the first read of the buffer, verify that each dquot is valid.
351 * We don't know what the id of the dquot is supposed to be, just that
352 * they should be increasing monotonically within the buffer. If the
353 * first id is corrupt, then it will fail on the second dquot in the
354 * buffer so corruptions could point to the wrong dquot in this case.
356 for (i
= 0; i
< ndquots
; i
++) {
357 struct xfs_disk_dquot
*ddq
;
360 ddq
= &d
[i
].dd_diskdq
;
363 id
= be32_to_cpu(ddq
->d_id
);
365 error
= xfs_qm_dqcheck(mp
, ddq
, id
+ i
, 0, XFS_QMOPT_DOWARN
,
366 "xfs_dquot_buf_verify");
374 xfs_dquot_buf_read_verify(
377 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
379 if (!xfs_dquot_buf_verify_crc(mp
, bp
) || !xfs_dquot_buf_verify(mp
, bp
)) {
380 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
381 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
386 * we don't calculate the CRC here as that is done when the dquot is flushed to
387 * the buffer after the update is done. This ensures that the dquot in the
388 * buffer always has an up-to-date CRC value.
391 xfs_dquot_buf_write_verify(
394 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
396 if (!xfs_dquot_buf_verify(mp
, bp
)) {
397 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
398 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
403 const struct xfs_buf_ops xfs_dquot_buf_ops
= {
404 .verify_read
= xfs_dquot_buf_read_verify
,
405 .verify_write
= xfs_dquot_buf_write_verify
,
409 * Allocate a block and fill it with dquots.
410 * This is called when the bmapi finds a hole.
418 xfs_fileoff_t offset_fsb
,
421 xfs_fsblock_t firstblock
;
422 xfs_bmap_free_t flist
;
424 int nmaps
, error
, committed
;
426 xfs_trans_t
*tp
= *tpp
;
430 trace_xfs_dqalloc(dqp
);
433 * Initialize the bmap freelist prior to calling bmapi code.
435 xfs_bmap_init(&flist
, &firstblock
);
436 xfs_ilock(quotip
, XFS_ILOCK_EXCL
);
438 * Return if this type of quotas is turned off while we didn't
441 if (!xfs_this_quota_on(dqp
->q_mount
, dqp
->dq_flags
)) {
442 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
446 xfs_trans_ijoin(tp
, quotip
, XFS_ILOCK_EXCL
);
448 error
= xfs_bmapi_write(tp
, quotip
, offset_fsb
,
449 XFS_DQUOT_CLUSTER_SIZE_FSB
, XFS_BMAPI_METADATA
,
450 &firstblock
, XFS_QM_DQALLOC_SPACE_RES(mp
),
451 &map
, &nmaps
, &flist
);
454 ASSERT(map
.br_blockcount
== XFS_DQUOT_CLUSTER_SIZE_FSB
);
456 ASSERT((map
.br_startblock
!= DELAYSTARTBLOCK
) &&
457 (map
.br_startblock
!= HOLESTARTBLOCK
));
460 * Keep track of the blkno to save a lookup later
462 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
464 /* now we can just get the buffer (there's nothing to read yet) */
465 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
467 mp
->m_quotainfo
->qi_dqchunklen
,
470 error
= xfs_buf_geterror(bp
);
473 bp
->b_ops
= &xfs_dquot_buf_ops
;
476 * Make a chunk of dquots out of this buffer and log
479 xfs_qm_init_dquot_blk(tp
, mp
, be32_to_cpu(dqp
->q_core
.d_id
),
480 dqp
->dq_flags
& XFS_DQ_ALLTYPES
, bp
);
483 * xfs_bmap_finish() may commit the current transaction and
484 * start a second transaction if the freelist is not empty.
486 * Since we still want to modify this buffer, we need to
487 * ensure that the buffer is not released on commit of
488 * the first transaction and ensure the buffer is added to the
489 * second transaction.
491 * If there is only one transaction then don't stop the buffer
492 * from being released when it commits later on.
495 xfs_trans_bhold(tp
, bp
);
497 if ((error
= xfs_bmap_finish(tpp
, &flist
, &committed
))) {
503 xfs_trans_bjoin(tp
, bp
);
505 xfs_trans_bhold_release(tp
, bp
);
512 xfs_bmap_cancel(&flist
);
514 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
520 struct xfs_mount
*mp
,
521 struct xfs_trans
*tp
,
522 struct xfs_dquot
*dqp
,
524 struct xfs_buf
**bpp
)
527 struct xfs_disk_dquot
*ddq
;
532 * Read the buffer without verification so we get the corrupted
533 * buffer returned to us. make sure we verify it on write, though.
535 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, dqp
->q_blkno
,
536 mp
->m_quotainfo
->qi_dqchunklen
,
540 ASSERT(*bpp
== NULL
);
541 return XFS_ERROR(error
);
543 (*bpp
)->b_ops
= &xfs_dquot_buf_ops
;
545 ASSERT(xfs_buf_islocked(*bpp
));
546 d
= (struct xfs_dqblk
*)(*bpp
)->b_addr
;
548 /* Do the actual repair of dquots in this buffer */
549 for (i
= 0; i
< mp
->m_quotainfo
->qi_dqperchunk
; i
++) {
550 ddq
= &d
[i
].dd_diskdq
;
551 error
= xfs_qm_dqcheck(mp
, ddq
, firstid
+ i
,
552 dqp
->dq_flags
& XFS_DQ_ALLTYPES
,
553 XFS_QMOPT_DQREPAIR
, "xfs_qm_dqrepair");
555 /* repair failed, we're screwed */
556 xfs_trans_brelse(tp
, *bpp
);
557 return XFS_ERROR(EIO
);
565 * Maps a dquot to the buffer containing its on-disk version.
566 * This returns a ptr to the buffer containing the on-disk dquot
567 * in the bpp param, and a ptr to the on-disk dquot within that buffer
573 xfs_disk_dquot_t
**O_ddpp
,
577 struct xfs_bmbt_irec map
;
578 int nmaps
= 1, error
;
580 struct xfs_inode
*quotip
= xfs_dq_to_quota_inode(dqp
);
581 struct xfs_mount
*mp
= dqp
->q_mount
;
582 xfs_dqid_t id
= be32_to_cpu(dqp
->q_core
.d_id
);
583 struct xfs_trans
*tp
= (tpp
? *tpp
: NULL
);
585 dqp
->q_fileoffset
= (xfs_fileoff_t
)id
/ mp
->m_quotainfo
->qi_dqperchunk
;
587 xfs_ilock(quotip
, XFS_ILOCK_SHARED
);
588 if (!xfs_this_quota_on(dqp
->q_mount
, dqp
->dq_flags
)) {
590 * Return if this type of quotas is turned off while we
591 * didn't have the quota inode lock.
593 xfs_iunlock(quotip
, XFS_ILOCK_SHARED
);
598 * Find the block map; no allocations yet
600 error
= xfs_bmapi_read(quotip
, dqp
->q_fileoffset
,
601 XFS_DQUOT_CLUSTER_SIZE_FSB
, &map
, &nmaps
, 0);
603 xfs_iunlock(quotip
, XFS_ILOCK_SHARED
);
608 ASSERT(map
.br_blockcount
== 1);
611 * Offset of dquot in the (fixed sized) dquot chunk.
613 dqp
->q_bufoffset
= (id
% mp
->m_quotainfo
->qi_dqperchunk
) *
616 ASSERT(map
.br_startblock
!= DELAYSTARTBLOCK
);
617 if (map
.br_startblock
== HOLESTARTBLOCK
) {
619 * We don't allocate unless we're asked to
621 if (!(flags
& XFS_QMOPT_DQALLOC
))
625 error
= xfs_qm_dqalloc(tpp
, mp
, dqp
, quotip
,
626 dqp
->q_fileoffset
, &bp
);
631 trace_xfs_dqtobp_read(dqp
);
634 * store the blkno etc so that we don't have to do the
635 * mapping all the time
637 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
639 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
641 mp
->m_quotainfo
->qi_dqchunklen
,
642 0, &bp
, &xfs_dquot_buf_ops
);
644 if (error
== EFSCORRUPTED
&& (flags
& XFS_QMOPT_DQREPAIR
)) {
645 xfs_dqid_t firstid
= (xfs_dqid_t
)map
.br_startoff
*
646 mp
->m_quotainfo
->qi_dqperchunk
;
648 error
= xfs_qm_dqrepair(mp
, tp
, dqp
, firstid
, &bp
);
653 return XFS_ERROR(error
);
657 ASSERT(xfs_buf_islocked(bp
));
659 *O_ddpp
= bp
->b_addr
+ dqp
->q_bufoffset
;
666 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
667 * and release the buffer immediately.
669 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
673 struct xfs_mount
*mp
,
677 struct xfs_dquot
**O_dqpp
)
679 struct xfs_dquot
*dqp
;
680 struct xfs_disk_dquot
*ddqp
;
682 struct xfs_trans
*tp
= NULL
;
687 dqp
= kmem_zone_zalloc(xfs_qm_dqzone
, KM_SLEEP
);
689 dqp
->dq_flags
= type
;
690 dqp
->q_core
.d_id
= cpu_to_be32(id
);
692 INIT_LIST_HEAD(&dqp
->q_lru
);
693 mutex_init(&dqp
->q_qlock
);
694 init_waitqueue_head(&dqp
->q_pinwait
);
697 * Because we want to use a counting completion, complete
698 * the flush completion once to allow a single access to
699 * the flush completion without blocking.
701 init_completion(&dqp
->q_flush
);
702 complete(&dqp
->q_flush
);
705 * Make sure group quotas have a different lock class than user
710 /* uses the default lock class */
713 lockdep_set_class(&dqp
->q_qlock
, &xfs_dquot_group_class
);
716 lockdep_set_class(&dqp
->q_qlock
, &xfs_dquot_project_class
);
723 XFS_STATS_INC(xs_qm_dquot
);
725 trace_xfs_dqread(dqp
);
727 if (flags
& XFS_QMOPT_DQALLOC
) {
728 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_DQALLOC
);
729 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_attrsetm
,
730 XFS_QM_DQALLOC_SPACE_RES(mp
), 0);
733 cancelflags
= XFS_TRANS_RELEASE_LOG_RES
;
737 * get a pointer to the on-disk dquot and the buffer containing it
738 * dqp already knows its own type (GROUP/USER).
740 error
= xfs_qm_dqtobp(&tp
, dqp
, &ddqp
, &bp
, flags
);
743 * This can happen if quotas got turned off (ESRCH),
744 * or if the dquot didn't exist on disk and we ask to
747 trace_xfs_dqread_fail(dqp
);
748 cancelflags
|= XFS_TRANS_ABORT
;
752 /* copy everything from disk dquot to the incore dquot */
753 memcpy(&dqp
->q_core
, ddqp
, sizeof(xfs_disk_dquot_t
));
754 xfs_qm_dquot_logitem_init(dqp
);
757 * Reservation counters are defined as reservation plus current usage
758 * to avoid having to add every time.
760 dqp
->q_res_bcount
= be64_to_cpu(ddqp
->d_bcount
);
761 dqp
->q_res_icount
= be64_to_cpu(ddqp
->d_icount
);
762 dqp
->q_res_rtbcount
= be64_to_cpu(ddqp
->d_rtbcount
);
764 /* initialize the dquot speculative prealloc thresholds */
765 xfs_dquot_set_prealloc_limits(dqp
);
767 /* Mark the buf so that this will stay incore a little longer */
768 xfs_buf_set_ref(bp
, XFS_DQUOT_REF
);
771 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
772 * So we need to release with xfs_trans_brelse().
773 * The strategy here is identical to that of inodes; we lock
774 * the dquot in xfs_qm_dqget() before making it accessible to
775 * others. This is because dquots, like inodes, need a good level of
776 * concurrency, and we don't want to take locks on the entire buffers
777 * for dquot accesses.
778 * Note also that the dquot buffer may even be dirty at this point, if
779 * this particular dquot was repaired. We still aren't afraid to
780 * brelse it because we have the changes incore.
782 ASSERT(xfs_buf_islocked(bp
));
783 xfs_trans_brelse(tp
, bp
);
786 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
796 xfs_trans_cancel(tp
, cancelflags
);
798 xfs_qm_dqdestroy(dqp
);
804 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
805 * a locked dquot, doing an allocation (if requested) as needed.
806 * When both an inode and an id are given, the inode's id takes precedence.
807 * That is, if the id changes while we don't hold the ilock inside this
808 * function, the new dquot is returned, not necessarily the one requested
809 * in the id argument.
814 xfs_inode_t
*ip
, /* locked inode (optional) */
815 xfs_dqid_t id
, /* uid/projid/gid depending on type */
816 uint type
, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
817 uint flags
, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
818 xfs_dquot_t
**O_dqpp
) /* OUT : locked incore dquot */
820 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
821 struct radix_tree_root
*tree
= xfs_dquot_tree(qi
, type
);
822 struct xfs_dquot
*dqp
;
825 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
826 if ((! XFS_IS_UQUOTA_ON(mp
) && type
== XFS_DQ_USER
) ||
827 (! XFS_IS_PQUOTA_ON(mp
) && type
== XFS_DQ_PROJ
) ||
828 (! XFS_IS_GQUOTA_ON(mp
) && type
== XFS_DQ_GROUP
)) {
833 if (xfs_do_dqerror
) {
834 if ((xfs_dqerror_target
== mp
->m_ddev_targp
) &&
835 (xfs_dqreq_num
++ % xfs_dqerror_mod
) == 0) {
836 xfs_debug(mp
, "Returning error in dqget");
841 ASSERT(type
== XFS_DQ_USER
||
842 type
== XFS_DQ_PROJ
||
843 type
== XFS_DQ_GROUP
);
845 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
846 ASSERT(xfs_inode_dquot(ip
, type
) == NULL
);
851 mutex_lock(&qi
->qi_tree_lock
);
852 dqp
= radix_tree_lookup(tree
, id
);
855 if (dqp
->dq_flags
& XFS_DQ_FREEING
) {
857 mutex_unlock(&qi
->qi_tree_lock
);
858 trace_xfs_dqget_freeing(dqp
);
864 mutex_unlock(&qi
->qi_tree_lock
);
866 trace_xfs_dqget_hit(dqp
);
867 XFS_STATS_INC(xs_qm_dqcachehits
);
871 mutex_unlock(&qi
->qi_tree_lock
);
872 XFS_STATS_INC(xs_qm_dqcachemisses
);
875 * Dquot cache miss. We don't want to keep the inode lock across
876 * a (potential) disk read. Also we don't want to deal with the lock
877 * ordering between quotainode and this inode. OTOH, dropping the inode
878 * lock here means dealing with a chown that can happen before
879 * we re-acquire the lock.
882 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
884 error
= xfs_qm_dqread(mp
, id
, type
, flags
, &dqp
);
887 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
894 * A dquot could be attached to this inode by now, since
895 * we had dropped the ilock.
897 if (xfs_this_quota_on(mp
, type
)) {
898 struct xfs_dquot
*dqp1
;
900 dqp1
= xfs_inode_dquot(ip
, type
);
902 xfs_qm_dqdestroy(dqp
);
908 /* inode stays locked on return */
909 xfs_qm_dqdestroy(dqp
);
910 return XFS_ERROR(ESRCH
);
914 mutex_lock(&qi
->qi_tree_lock
);
915 error
= -radix_tree_insert(tree
, id
, dqp
);
916 if (unlikely(error
)) {
917 WARN_ON(error
!= EEXIST
);
920 * Duplicate found. Just throw away the new dquot and start
923 mutex_unlock(&qi
->qi_tree_lock
);
924 trace_xfs_dqget_dup(dqp
);
925 xfs_qm_dqdestroy(dqp
);
926 XFS_STATS_INC(xs_qm_dquot_dups
);
931 * We return a locked dquot to the caller, with a reference taken
937 mutex_unlock(&qi
->qi_tree_lock
);
940 ASSERT((ip
== NULL
) || xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
941 trace_xfs_dqget_miss(dqp
);
949 struct xfs_dquot
*dqp
)
951 struct xfs_quotainfo
*qi
= dqp
->q_mount
->m_quotainfo
;
952 struct xfs_dquot
*gdqp
;
953 struct xfs_dquot
*pdqp
;
955 trace_xfs_dqput_free(dqp
);
957 if (list_lru_add(&qi
->qi_lru
, &dqp
->q_lru
))
958 XFS_STATS_INC(xs_qm_dquot_unused
);
961 * If we just added a udquot to the freelist, then we want to release
962 * the gdquot/pdquot reference that it (probably) has. Otherwise it'll
963 * keep the gdquot/pdquot from getting reclaimed.
965 gdqp
= dqp
->q_gdquot
;
968 dqp
->q_gdquot
= NULL
;
971 pdqp
= dqp
->q_pdquot
;
974 dqp
->q_pdquot
= NULL
;
979 * If we had a group/project quota hint, release it now.
988 * Release a reference to the dquot (decrement ref-count) and unlock it.
990 * If there is a group quota attached to this dquot, carefully release that
991 * too without tripping over deadlocks'n'stuff.
995 struct xfs_dquot
*dqp
)
997 ASSERT(dqp
->q_nrefs
> 0);
998 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1000 trace_xfs_dqput(dqp
);
1002 if (--dqp
->q_nrefs
> 0)
1005 xfs_qm_dqput_final(dqp
);
1009 * Release a dquot. Flush it if dirty, then dqput() it.
1010 * dquot must not be locked.
1019 trace_xfs_dqrele(dqp
);
1023 * We don't care to flush it if the dquot is dirty here.
1024 * That will create stutters that we want to avoid.
1025 * Instead we do a delayed write when we try to reclaim
1026 * a dirty dquot. Also xfs_sync will take part of the burden...
1032 * This is the dquot flushing I/O completion routine. It is called
1033 * from interrupt level when the buffer containing the dquot is
1034 * flushed to disk. It is responsible for removing the dquot logitem
1035 * from the AIL if it has not been re-logged, and unlocking the dquot's
1036 * flush lock. This behavior is very similar to that of inodes..
1039 xfs_qm_dqflush_done(
1041 struct xfs_log_item
*lip
)
1043 xfs_dq_logitem_t
*qip
= (struct xfs_dq_logitem
*)lip
;
1044 xfs_dquot_t
*dqp
= qip
->qli_dquot
;
1045 struct xfs_ail
*ailp
= lip
->li_ailp
;
1048 * We only want to pull the item from the AIL if its
1049 * location in the log has not changed since we started the flush.
1050 * Thus, we only bother if the dquot's lsn has
1051 * not changed. First we check the lsn outside the lock
1052 * since it's cheaper, and then we recheck while
1053 * holding the lock before removing the dquot from the AIL.
1055 if ((lip
->li_flags
& XFS_LI_IN_AIL
) &&
1056 lip
->li_lsn
== qip
->qli_flush_lsn
) {
1058 /* xfs_trans_ail_delete() drops the AIL lock. */
1059 spin_lock(&ailp
->xa_lock
);
1060 if (lip
->li_lsn
== qip
->qli_flush_lsn
)
1061 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_CORRUPT_INCORE
);
1063 spin_unlock(&ailp
->xa_lock
);
1067 * Release the dq's flush lock since we're done with it.
1073 * Write a modified dquot to disk.
1074 * The dquot must be locked and the flush lock too taken by caller.
1075 * The flush lock will not be unlocked until the dquot reaches the disk,
1076 * but the dquot is free to be unlocked and modified by the caller
1077 * in the interim. Dquot is still locked on return. This behavior is
1078 * identical to that of inodes.
1082 struct xfs_dquot
*dqp
,
1083 struct xfs_buf
**bpp
)
1085 struct xfs_mount
*mp
= dqp
->q_mount
;
1087 struct xfs_disk_dquot
*ddqp
;
1090 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1091 ASSERT(!completion_done(&dqp
->q_flush
));
1093 trace_xfs_dqflush(dqp
);
1097 xfs_qm_dqunpin_wait(dqp
);
1100 * This may have been unpinned because the filesystem is shutting
1101 * down forcibly. If that's the case we must not write this dquot
1102 * to disk, because the log record didn't make it to disk.
1104 * We also have to remove the log item from the AIL in this case,
1105 * as we wait for an emptry AIL as part of the unmount process.
1107 if (XFS_FORCED_SHUTDOWN(mp
)) {
1108 struct xfs_log_item
*lip
= &dqp
->q_logitem
.qli_item
;
1109 dqp
->dq_flags
&= ~XFS_DQ_DIRTY
;
1111 spin_lock(&mp
->m_ail
->xa_lock
);
1112 if (lip
->li_flags
& XFS_LI_IN_AIL
)
1113 xfs_trans_ail_delete(mp
->m_ail
, lip
,
1114 SHUTDOWN_CORRUPT_INCORE
);
1116 spin_unlock(&mp
->m_ail
->xa_lock
);
1117 error
= XFS_ERROR(EIO
);
1122 * Get the buffer containing the on-disk dquot
1124 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dqp
->q_blkno
,
1125 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
, NULL
);
1130 * Calculate the location of the dquot inside the buffer.
1132 ddqp
= bp
->b_addr
+ dqp
->q_bufoffset
;
1135 * A simple sanity check in case we got a corrupted dquot..
1137 error
= xfs_qm_dqcheck(mp
, &dqp
->q_core
, be32_to_cpu(ddqp
->d_id
), 0,
1138 XFS_QMOPT_DOWARN
, "dqflush (incore copy)");
1142 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1143 return XFS_ERROR(EIO
);
1146 /* This is the only portion of data that needs to persist */
1147 memcpy(ddqp
, &dqp
->q_core
, sizeof(xfs_disk_dquot_t
));
1150 * Clear the dirty field and remember the flush lsn for later use.
1152 dqp
->dq_flags
&= ~XFS_DQ_DIRTY
;
1154 xfs_trans_ail_copy_lsn(mp
->m_ail
, &dqp
->q_logitem
.qli_flush_lsn
,
1155 &dqp
->q_logitem
.qli_item
.li_lsn
);
1158 * copy the lsn into the on-disk dquot now while we have the in memory
1159 * dquot here. This can't be done later in the write verifier as we
1160 * can't get access to the log item at that point in time.
1162 * We also calculate the CRC here so that the on-disk dquot in the
1163 * buffer always has a valid CRC. This ensures there is no possibility
1164 * of a dquot without an up-to-date CRC getting to disk.
1166 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
1167 struct xfs_dqblk
*dqb
= (struct xfs_dqblk
*)ddqp
;
1169 dqb
->dd_lsn
= cpu_to_be64(dqp
->q_logitem
.qli_item
.li_lsn
);
1170 xfs_update_cksum((char *)dqb
, sizeof(struct xfs_dqblk
),
1175 * Attach an iodone routine so that we can remove this dquot from the
1176 * AIL and release the flush lock once the dquot is synced to disk.
1178 xfs_buf_attach_iodone(bp
, xfs_qm_dqflush_done
,
1179 &dqp
->q_logitem
.qli_item
);
1182 * If the buffer is pinned then push on the log so we won't
1183 * get stuck waiting in the write for too long.
1185 if (xfs_buf_ispinned(bp
)) {
1186 trace_xfs_dqflush_force(dqp
);
1187 xfs_log_force(mp
, 0);
1190 trace_xfs_dqflush_done(dqp
);
1196 return XFS_ERROR(EIO
);
1200 * Lock two xfs_dquot structures.
1202 * To avoid deadlocks we always lock the quota structure with
1203 * the lowerd id first.
1212 if (be32_to_cpu(d1
->q_core
.d_id
) >
1213 be32_to_cpu(d2
->q_core
.d_id
)) {
1214 mutex_lock(&d2
->q_qlock
);
1215 mutex_lock_nested(&d1
->q_qlock
, XFS_QLOCK_NESTED
);
1217 mutex_lock(&d1
->q_qlock
);
1218 mutex_lock_nested(&d2
->q_qlock
, XFS_QLOCK_NESTED
);
1221 mutex_lock(&d1
->q_qlock
);
1223 mutex_lock(&d2
->q_qlock
);
1231 kmem_zone_init(sizeof(struct xfs_dquot
), "xfs_dquot");
1236 kmem_zone_init(sizeof(struct xfs_dquot_acct
), "xfs_dqtrx");
1237 if (!xfs_qm_dqtrxzone
)
1238 goto out_free_dqzone
;
1243 kmem_zone_destroy(xfs_qm_dqzone
);
1251 kmem_zone_destroy(xfs_qm_dqtrxzone
);
1252 kmem_zone_destroy(xfs_qm_dqzone
);